Overview

AgentScope Memory Finance Bot is a local-first AI finance assistant that combines:

• Persistent per-user memory stored in SQLite
• Local LLM inference using Ollama (Llama 3.2)
• Live market data via yfinance
• Tool-augmented reasoning using AgentScope ReActAgent

This project is designed for fully local execution (no OpenAI dependency).

---

Features

• 🧠 Persistent memory per user (memories/<user_id>.db)
• 🏦 Stock market quotes via yfinance
• 🤖 Local LLM via Ollama (llama3.2:3b)
• 🔧 Tool-based reasoning (ReAct agent architecture)
• ⚡ Async memory + engine lifecycle management

---

Project Structure

.
|-- main.py
|-- pyproject.toml
|-- uv.lock
|-- .env (optional, not required for LLM)
`-- memories/      # auto-created at runtime

---

Local Setup (IMPORTANT)

1. Prerequisites

Install:

• Python 3.12+
• Ollama

Check Python:

python --version

---

2. Install Ollama (Required)

Install Ollama: https://ollama.com/download

Then pull the model:

ollama pull llama3.2:3b

Start server:

ollama serve

Ensure it runs on:

http://localhost:11434

---

3. Create project

git clone https://github.com/pawan941394/agentscope-memory-finance-bot.git
cd agentscope-memory-finance-bot

or

mkdir agentscope-memory-finance-bot
cd agentscope-memory-finance-bot

---

4. Virtual environment

python -m venv .venv
source .venv/bin/activate   # mac/linux
.venv\Scripts\activate      # windows

---

5. Install dependencies

pip install agentscope sqlalchemy aiosqlite yfinance

or

uv sync

---

Run the app

python main.py

You will see:

Enter user id:

Use same user_id to continue memory across sessions.

---

How it works (Architecture)

• Agent: AgentScope ReActAgent
• Model: Ollama (llama3.2:3b)
• Memory: AsyncSQLAlchemyMemory (SQLite per user)
• Tool: Yahoo Finance via yfinance

---

Tool: Stock Quotes

Example tool input:

AAPL, MSFT, TSLA

Output includes:

• Price
• Change
• Volume
• Day high/low

---

Example Prompts

• “Get AAPL, MSFT and summarize market movement”
• “Compare NVDA vs AMD performance”
• “What do you remember about my investing style?”

---

Memory Design

• Stored in: memories/<sanitized_user_id>.db
• Backend: SQLite (async)
• Session: default_session

---

Notes

• Fully local LLM (no OpenAI API required)
• Market data is via yfinance (may be delayed)
• Designed for learning agent systems + tool use

---

Troubleshooting

Ollama not responding

ollama serve

Model missing

ollama pull llama3.2:3b

Import errors

pip install agentscope sqlalchemy aiosqlite yfinance

---

Roadmap

• Streamlit UI/FE
• Multi-agent finance assistant

---

Contributing

PRs welcome. Keep it simple, local-first, and modular.

---

License

MIT